|
| 基于Sentinel时间序列的玉麦轮作区识别方法 |
| Identification method of corn-wheat rotation area based on Sentinel time series |
| 投稿时间：2024-11-17 |
| DOI：10.13254/j.jare.2024.0901 |
| 中文关键词: 时间序列 Google Earth Engine 随机森林 轮作区 哨兵数据 冬小麦 夏玉米 |
| 英文关键词: time series Google Earth engine random forest rotation area Sentinel data winter wheat summer corn |
| 基金项目:山东省自然科学基金项目（ZR2021MD096） |
|
| 摘要点击次数: 367 |
| 全文下载次数: 242 |
| 中文摘要: |
| 为探究利用遥感技术识别不同作物轮作区，准确获取大规模冬小麦-夏玉米轮作区的面积与种植区域的分布情况，通过分析Sentinel-2数据上作物反射率的时空差异，构建了轮作植被指数（CRVI），以刻画近似物候作物在时间尺度上的异质性，结合Sentinel-1数据构建时间序列并进行优化，设置四种分类特征方案，采用随机森林算法提取冬小麦-夏玉米轮作区，并对分类结果进行精度评估。结果表明：冬小麦-夏玉米种植区主要分布在菏泽、济宁两地的黄河冲积平原地区，枣庄、临沂种植区分布较少且零散。不同方案分类结果在大范围种植区域分类结果相似，在地块破碎地区重要性较高特征（方案四）更加准确，但其错分概率较大，总体来说时间序列分类结果符合实际。方案一分类结果总体精度为97.68%，Kappa系数为0.977 6，其中，冬小麦-夏玉米遥感提取总体面积与统计年鉴数据相比，准确度为 93.50%。研究表明，通过引入 CRVI，并结合 Sentinel-1与 Sentinel-2时间序列数据，实现了高精度的大规模冬小麦-夏玉米轮作区识别。该方法不仅提高了分类精度，还为农业生产提供了可靠的数据支持，展示了在复杂种植结构下作物分类的新思路和技术手段。 |
| 英文摘要: |
| To investigate the use of remote sensing technology in identifying different crop rotation areas and accurately obtaining the area and distribution of large-scale winter wheat-summer maize rotation regions, a Crop Rotation Vegetation Index（CRVI）was developed by analyzing the spatio-temporal differences in crop reflectivity from Sentinel-2 data. This index captured the heterogeneity of crops with similar phenologies across time scales. Time series constructed from Sentinel-1 data were optimized, and four classification feature schemes were set up. The random forest algorithm was employed to extract winter wheat-summer maize rotation areas, and the classification results were evaluated for accuracy. The results indicated that the winter wheat-summer maize planting areas were mainly distributed in the Yellow River alluvial plains of Heze and Jining, while the planting areas in Zaozhuang and Linyi were sparse and scattered. The classification results of different schemes were similar in large-scale planting areas, and features with higher importance in fragmented fields were more accurate, albeit with a higher misclassification probability. Overall, the time series classification results were consistent with reality. The overall accuracy of the final scheme one classification result was 97.68%, and the Kappa coefficient was 0.977 6. Specifically, when compared with statistical yearbook data, the overall accuracy of remotely sensed extraction of winter wheat-summer maize areas was 93.50%. The study showed that by introducing the CRVI and combining Sentinel-1 and Sentinel-2 time series data, highaccuracy identification of large-scale winter wheat-summer maize rotation areas could be achieved. This method not only improveds classification accuracy but also provideds reliable data support for agricultural production, demonstrating a new approach and technical means for crop classification under complex planting structures. |
| HTML
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
